Research On Image Feature Extraction Based On Deep Learning And Visual Servoing

The design and selection of features is one of the hotspots and key technologies in the field of robot visual servoing research.It is a combination of computer vision and robot servoing control.In recent years,deep learning has been successfully applied in the field of computer vision.The image feature extraction based on deep learning is an important research direction.This paper take visual servoing as background,propose a feature extraction method based on deep learning to solve the problem of feature design in visual servoing.To solve the problem of model dependence in visual servoing,the feature extraction method proposed in this paper is applied,and the Kalman filter method is applied to the online estimation of the system model to implement the calibration-free servo tracking control.The main work of this article includes:(1)A keypoint detection method based on deep learning is proposed.The method consists of constructing keypoint data sets,designing multi-scale keypoint detection networks,and designing keypoint objective functions.This method combines deep convolutional networks and multi-scale recursive convolutional networks.Compared with other methods,it can obtain more repeatable keypoints to cope with the changing scenes in visual servoing.(2)A keypoint feature extraction method based on deep learning is proposed.This method can extract local features of keypoint for matching.This method uses a convolutional network structure,pre-train the convolutional network based on Siamese Network Structure firstly,and then adjusted to Triplet Tetwork Structure to improve the accuracy.The matching results with other methods show that the features based on deep learning are better than traditional features and are more suitable for visual servoing tasks.(3)To solve the model dependent problem in visual servoing,the Kalman filter method is used to estimate the image Jacobian matrix online,and the motion in the robot's task space is mapped to the image feature space.Then the robot control quantity is directly calculated based on the image feature error.The simulation results of binocular uncalibrated visual servoing system based on this strategy show that the method has good dynamic characteristics and small steady-state error,which meets the requirements of non-calibration visual servoing control.(4)Based on the industrial robot platform and industrial camera,a binocular visual servoing system was constructed,and the proposed visual feature extraction method and noncalibrated visual servoing control strategy were applied to realize the calibration-free visual servoing tracking.Experimental results verify the effectiveness of the proposed method.